Techniques for robust planning in degradable multiagent systems

Abstract

We discussed the issue of planning in a degradable multiagent system and presented a framework in which FT techniques can be integrated into agent planning/coordination to handle uncertainty in domain problem solving as well as organizational change. Our framework starts at the decision-theoretic level to formally define performability in multiagent problem solving, then moves on to the agent planning and coordination level, and finally at the organizational level. Our ultimate goal, evidently, is to create fault-tolerant multiagent systems, and to allow simple translation or application of the many FT techniques that have been design for traditional systems in multiagent systems. The real challenge, is not about applying a few techniques that address the reliability concern of some aspects of the system, but to incorporate robustness into every aspect of the system design. At this point, a lot of work remains to be done in order to prove that those mechanisms indeed work and that the framework does allow easy integration of FT techniques in MAS. Another important aim of this research is to enhance and extend existing planning frameworks-so that these planning frameworks can still find their applications, but with the proposed mechanisms they can implement some FT mechanisms and also form a large solution by piecing together smaller solutions. In both cases, coordination is really the key toward enhancing the robustness of the system-to handle both task failures as well as organizational failures. The FT techniques involved in the proposed mechanisms are not new-they have been used in the context of distributed systems or fault-tolerant computing, but new meanings are being developed in the context of autonomous agent systems and robust organizations.